Year: 2026
Type: dissertation
Abstract: With recent advances in generative AI and conversational agent technologies, AI is increasingly expected to function not merely as an efficient tool, but as a "teammate" that collaborates with humans in task execution.For effective Collaborative Problem Solving (CPS) between humans and AI, not only high functional performance on the AI side but also humans' psychological acceptance of AI as a team member is essential.Prior research has shown that psychological acceptance is facilitated by perceptions of warmth.In studies on physically embodied robots, it has been suggested that physiological or bodily responses to task-irrelevant stimuli can evoke social presence and enhance perceived warmth.However, it remains insufficiently examined whether similar effects occur in text-based agents such as chatbots that lack a physical body.In particular, the effects of embodied-like responses that expose AI imperfection or "weakness" (pseudoembodiment) on human acceptance attitudes remain unclear.Therefore, this study aimed to empirically investigate whether a chatbot's pseudoembodied responses expressing "weakness" to task-irrelevant stimuli promote perceived warmth and psychological acceptance.An experimental decision-making task based on a modified version of the OECD PISA CPS task "The Visit" was conducted, in which participants collaborated with a chatbot to decide on a destination.A total of 50 participants (30 Japanese and 20 international students) took part in a between-subjects experiment.Participants were assigned either to a pseudo-embodied response condition, in which the chatbot displayed physiological reactions (increased heart rate) to taskirrelevant sounds, or to a non-response condition, in which the chatbot ignored the sounds and proceeded with the task.After the task, questionnaire surveys and interviews were conducted to measure perceived warmth, psychological acceptance, and potential confounding factors such as AI literacy.Independent-samples t-tests conducted on the full sample revealed no statistically significant differences between conditions in perceived warmth or psychological acceptance, and the overall hypothesis was not supported.However, correlation analyses showed a strong positive relationship between perceived warmth and psychological acceptance, supporting the association that higher warmth perception corresponds to higher acceptance.Furthermore, exploratory analyses focusing on cultural background yielded important findings.Among international students, the pseudo-embodied response condition resulted in significantly higher perceived warmth and psychological acceptance compared to the non-response condition, and mediation analysis confirmed a process in which warmth promoted psychological acceptance.In contrast, no positive effects of bodily responses were observed among Japanese participants; instead, such responses tended to be perceived negatively as noise that interfered with task performance.Text analysis of interview data further indicated that international students tended to interpret bodily responses positively as expressions of dialogue or emotion, whereas Japanese participants tended to interpret them functionally as mechanical malfunctions or unnecessary conversation.These results suggest that the effects of pseudo-embodied responses in chatbots strongly depend on users' cultural background and AI literacy.For international students, expressions of AI "weakness" functioned as cues of human-like social presence and contributed to reducing psychological distance.Conversely, among Japanese participants, strong functional expectations for accuracy and efficiency in decision-making tasks appeared to cause task-irrelevant emotional responses to hinder acceptance.Additionally, participants with higher AI literacy tended to interpret AI behavior more favorably and showed higher acceptance.This study empirically demonstrates that even chatbots without physical bodies can enhance perceived warmth and psychological acceptance among international users by presenting pseudo-embodiment in the form of responses to task-irrelevant stimuli.The findings indicate that designing AI as a teammate requires not only functional optimization but also careful adjustment of human-likeness according to users' cultural background and task context.While incorporating embodied expressions of "weakness" may be beneficial in global AI design, future research in Japan should explore forms of embodiment that do not interfere with task performance, such as visualization of internal states, as well as effects in long-term interactions.This study provides important implications for agent design in human-AI coexistence from the perspectives of cultural differences and psychological acceptance.
Author: 中村, 都夢
Language: Japanese
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7162975501